A Path Planning Method Based On Reinforcement Learning

A path is a curve or discrete sequence of points that connects the starting and ending points.The process of generating these curves and sequence points is called path planning.In general,path planning consists of two parts,the environment model and the path generation algorithm.The current focus on reinforcement learning is on enabling agents to perform various tasks in virtual environments.Researchers have been able to make agents perform well in video games,and perform far better than humans in complex game environments such as starcraft II.The model of training reinforcement learning in virtual environment also needs two elements--virtual environment and reinforcement learning algorithm,which is similar to the composition of path planning.this paper proposes a method of using reinforcement learning to solve the problem of path planning.In practical application scenarios,a planned route must have practical significance.For example,if a z-shaped route is planned for a vehicle,it is difficult for the vehicle to strictly follow the route due to its own kinetic characteristics,so the route has no practical application value.In order to plan out the path of the application in a real-world scenario,this paper proposes a method that consider two aspects of kinematics and kinetics of path planning.Then the path planning problem is converted into reinforcement learning problems,using openai open source of reinforcement learning algorithm,proximal policy optimization algorithm(PPO)training convolution neural network model.In this paper,the experiment is carried out on a tool machine and a fascinating result is obtained.